RerrFact
This repository contains the code for: RerrFact model for SciVer shared task.
Setup for Inference
1. Download SciFact database
Download the SciFact database from here.
2. Installing requirements
Install the requirements using the following command for abstract retrieval and rationale selection module.
pip install -r abstract,rationale_requirements.txt
Install the requirements using the following command for label prediction module.
pip install -r label_requirements.txt
3. Download trained models
Download the trained models using this link.
4. Using pre-trained models
Abstract Retrieval
python ./inference/abstract-retrieval.py \
--corpus ./data/corpus.jsonl \
--dataset ./data/claims_test.jsonl \
--model ./saved_models/abstract_retrieval_model_here \
--output ./prediction/abstract_retrieval_test_predictions.jsonl
Rationale Selection
python ./inference/rationale-selection.py \
--corpus ./data/corpus.jsonl \
--dataset ./data/claims_test.jsonl \
--abstract ./prediction/abstract_retrieval_test_predictions.jsonl \
--model ./saved_models/rationale_selection_model_here \
--output ./prediction/
Label Prediction
python inference/label-prediction.py \
--corpus '/data/corpus.jsonl' \
--dataset './data/claims_test.jsonl' \
--rationale-selection './prediction/rationale_selection.jsonl' \
--model_n './saved_models/neutral_classifer_here' \
--model_s './saved_models/support_classifier_here' \
--output './prediction/label_pred_test.jsonl'
Retrain models
Refer to training/Abstract-retrieval.ipynb
for training abstract retrieval module.
Refer to training/Rationale-selection.ipynb
for training rationale selection module.
Refer to training/Label-prediction.ipynb
for training label prediction module.